Titanium-Class CRPS Power for GPU-Dense Racks
From 550W classic CRPS to 5.5kW M-CRPS — 96% peak efficiency at 50% load, PMBus telemetry, and hot-swap N+1 redundancy. One supplier across the rack power chain.
One Ladder, Every Rack Segment
* Development samples, available for joint evaluation · ** Roadmap, offered as joint development with lead customers
Pick Your Power Segment
Classic CRPS Titanium
The workhorse segment for 1U/2U servers, storage and switches. CRPS-185 envelope, 12V main output, PMBus 1.2.
Explore series →
High-Density CRPS Titanium
Class-leading power density up to 75 W/in³ class, C19 input, active current sharing for N+1 arrays feeding GPU nodes.
Explore series →
M-CRPS Modular
Modular output section on the OCP M-CRPS path, targeting 54V ORv3 slot compatibility. Now in development samples.
Development status →
GPU Server Power
An 8-GPU HGX node draws 10kW or more. We pair CRPS arrays, 54V architectures and telemetry to keep accelerators fed.
See solutions →
Power Shelf 33kW
Six 5.5kW modules make a 33kW shelf with 5+1 redundancy. We are building toward it with lead partners.
Roadmap →
Built on Real Production Lines
SMT+DIP lines, automated ATE test, a burn-in aging room and in-house safety compliance — the four pillars behind every unit.
Our capabilities →
The Money Math of 96% Efficiency
Titanium moves the 50%-load efficiency bar from roughly 92% (Gold) to 96%. On a 10kW compute load that is about 430 watts of heat you never generate — every hour, on every node.
Across a 100-rack AI cluster, that difference compounds into megawatt-hours per year, smaller cooling budgets, and headroom you can spend on GPUs instead.
Uptime Institute puts the cost of data center downtime near $9,000 per hour. Redundant titanium-class supplies with active current sharing are the cheapest insurance in the rack.
How we reach 96%Four Disciplines Behind Every Watt
Titanium Efficiency
LLC resonant stages and synchronous rectification tuned for the 10/50/100% points that Titanium certifies.
Thermal Design
Continuous output from −5°C to +55°C, front or back airflow options, and NTC placements mapped in wind-tunnel tests.
PMBus Telemetry
Voltage, current, temperature and fan speed per rail — plus power throttling and cold-redundancy modes on request.
Testing & Reliability
ATE end-of-line test, 100% burn-in screening and MTBF targets above 250k hours at 40°C per Telcordia SR-332.
Four Pillars of Production
SMT + DIP mixed lines feed automated ATE stations; every unit finishes a burn-in cycle in our aging room before the safety lab signs off. Capacity grows with demand — we publish process, not promises.
Tour the FactoryWhere Our Power Goes
AI Data Centers
H100-class nodes draw 700W per GPU; B200 pushes 1000W. Feeding them is our design center.
Hyperscale & ODM Direct
Open architectures, ORv3 alignment and batch-to-batch consistency for ODM-direct procurement.
Enterprise Servers
Classic CRPS from 550W with the chassis compatibility that 1U/2U platforms demand.
Storage Systems
7×24 write-heavy loads, dual power paths and current sharing across 800W–2kW units.
Networking
ToR and spine switches: 550W–1300W with front or back airflow to match your aisle plan.
Telecom & Edge
−48Vdc legacy and 54V new builds, wide-temperature operation and hardened protection at the edge.
What an ODM-Ready Partner Looks Like
| TitanWatt | Typical brand-line PSU | |
|---|---|---|
| Custom output configuration | Connector layout, harness and bezel customizable from 100 pcs | Fixed SKUs, custom at 10k+ pcs |
| PMBus command tailoring | Command set adapted to your BMC at project start | Standard command set only |
| Certification support | IEC/UL 62368-1 CB, CE, UKCA, FCC, RoHS/REACH; regional certs per market plan | Base certs only |
| Engineering response | Direct FAE access, 48-hour reply commitment | Ticket queue |
| Batch traceability | Serial-level test data retained for every shipment | Batch-level at best |
Power Engineering, Explained
Titanium vs Gold: the Electricity Bill Math
Where the 4-point efficiency gap actually shows up on your power bill — and when it pays back.
6 min readCRPS vs M-CRPS: What Actually Changed
Same 73.5×40×185mm envelope, very different output philosophy. The modular split explained.
7 min readGPU Rack Power Trends and the 54V Move
From 5kW racks to 120kW AI pods — how power architectures are scaling, and what comes next.
8 min readFrequently Asked
Powering Your Next Rack?
Tell us your wattage, architecture and volumes. An engineer — not an autoresponder — replies within 48 hours.
Browse Products